Objective The urban governance of Beijing, the capital of China, not only affects Beijing’s development quality, but also directly impacts China’s international image. Therefore, in the face of the problem of a large amount of bare land generated in urban construction, the Beijing Municipal Government innovatively proposed the special work of “net uncovering for greening” in 2021, aiming to reduce dust pollution, improve the quality of Beijing’s ecological environment, and achieve the beautiful vision of “Green Beijing” by effectively managing the construction sites, temporary storage yards and other areas covered with green plastic cover net. However, in the process of implementing this special work, how to efficiently and accurately monitor the status of the green cover net has become an urgent technical problem to be solved. There is still a gap in research in this field both domestically and internationally, and there is an urgent need for a scientific and systematic monitoring system to support the effective promotion of the special work of “net uncovering for greening”.

Methods Taking the special work of “net uncovering for greening” in Beijing as an example, this research combines the technical means of remote sensing interpretation and deep learning with the, U-Net model to train the samples of tarpaulin cover. Then, through human-computer interaction mode, the extraction of the green tarpaulin cover and the land parcel under management is achieved. Specifically, the research first utilizes high-resolution satellite remote sensing images as the basic data source, and ensures data quality through preprocessing measures such as image correction and enhancement. Subsequently, representative samples of green tarpaulin cover are selected for annotation, and a training dataset is constructed. This process is crucial because high-quality sample data directly affects the accuracy and generalization capability of subsequent model training. The U-Net model, with its unique encoding decoding structure, is able to learn and capture the fine features of the green tarpaulin cover as shown in corresponding images, thereby achieving precise segmentation. Through training with a large number of samples, the U-Net model can gradually learn how to distinguish between green tarpaulin cover and other types of surface cover, such as bare soil and vegetation. In practical operation, in order to improve monitoring efficiency and accuracy, the research team also introduces a human-computer interaction mode. This mode allows professionals to verify and correct the preliminary segmentation results of the model, especially in complex scenes or areas with fuzzy edges. Manual intervention can significantly improve the reliability of the results. This mode can not only optimize the performance of the model, but also promotes seamless integration between the model and actual operations, making monitoring work more efficient and flexible.

Results This research constructs a complete working system of “establishment − implementation − verification − monitoring”. Specifically, “establishment” represents the establishment of annual accounting data and the clarification of annual work tasks; “implementation” refers to the implementation and feedback of the work of uncovering the net and promoting green development, namely the work of uncovering the net of classified land parcels; “verification” refers to the verification of uncovered land parcels. The remote sensing technology department verifys the implementation of uncovered land parcels based on feedback from the location information of uncovered land parcels and the description of uncovering situation provided by the implementation department, with high-resolution satellite images as the data source; “monitoring” refers to the dynamic monitoring of green tarpaulin cover, involving the reduction, addition and variation of green tarpaulin cover based on the background data of green tarpaulin cover. From 2020 to 2024, Beijing has completed 8 phases of city-wide monitoring of green tarpaulin cover net using the working system of “net uncovering for greening”, effectively grasping the distribution and dynamic changes of tarpaulin cover across the city, and providing scientific basis for government decision-making. At the same time, with the deepening of monitoring work, the research team has continuously expanded and improved the sample library of tarpaulin cover, with more types of tarpaulin cover materials, environmental conditions, and image data under seasonal changes being incorporated.

Conclusion In summary, the special work of “net uncovering for greening” in Beijing is not only an innovative practice of urban management, but also a successful case of deep integration of satellite remote sensing and deep learning technology. By establishing a scientific working system, efficient and accurate monitoring of land parcels subject to “net uncovering for greening” has been achieved, providing strong support for the ecological environment governance of mega cities. With the establishment and continuous enrichment of the sample library of tarpaulin cover, further support can be provided for intelligent interpretation of remote sensing images. In this research, the working system is expected to be promoted and applied in more fields, contributing to the construction of greener and smarter cities. Meanwhile, it also provides valuable experience and inspiration for similar work at home and abroad, thus upgrading urban management and environmental protection to a higher level.